Content deleted Content added
No edit summary |
|||
Line 303:
[[user:Masparasol|Masparasol]] Oct 22, 2005
===A way forward?===
Line 708 ⟶ 709:
The edits and comments regarding financial astrology are inappropriate and unsourced. It's true that the [[Market Technicians Association]] has not defined technical analysis, but it clearly does '''not''' include financial astrology in the [http://www.nyif.com/courses/exam_2011.html CMT course curriculum].
--[[User:Rgfolsom|Rgfolsom]] 17:39, 8 June 2007 (UTC)
== Neural Networks ==
I've added 2 references to this, the one to Funahashi and the one to Hornik in the same journal (and volume). The Hornik article is slightly later and is probably the first source of the phrase 'universal function approximators.' The Funahashi article refers only to continuous functions, and says that for function defined on a bounded region there is an arbitrarily good uniform approximation using a 3 level neural network allowing large numbers of neurons in the middle layer.
Also it would be good if someone had an idea how this section fits into the article.
Just as 'technical analysis' is controversial, also the use of neural networks in attempting to extrapolate a single function of time is also controversial in the same way the usefulness of Fibonacci numbers or Elliot Waves is controversial. For the 1 or 2 articles finding a statistically significant gain, many more do not. But the concept of approximating a function is not controversial and neural networks perhaps fit in in the sense that they bridge the gap between fundamental analysis where actual equations can be written down, and there is no question effects exist, on the one hand, versus technical analysis which is intuition and in some opinions questionable. If a function of several variables can be empirically approximated, it can be used even if the precise equation is not known or understood. [[User:82.26.87.140|82.26.87.140]] 23:48, 8 June 2007 (UTC)
| |||